Wrathemperor/SkinCancerDetection_HAM100000 ? reverse-engineered prompt

Build me a skin lesion classification app for the HAM10000 dataset that can train a model and then let a user upload an image in a simple web page to get an instant prediction and confidence score.

I want the project to include the full pipeline, starting with preparing the dataset, splitting it into train, validation, and test sets, and handling the class imbalance by oversampling the rare classes. Use image augmentation during training, then train a transfer learning model based on MobileNetV3Small for 7 class classification. Save the best model, show training history and a confusion matrix, and keep the results in a folder.

After training, the app should run as a Flask website with a clean upload form and a results page that shows the predicted lesion type. Please also include a clear medical disclaimer saying it is only for educational use, not a real diagnosis tool. If you need to check current docs or best practices for any library, look them up online.